基于YOLOv5的司机状态检测系统研究

Study on Driver Status Detection System Based on YOLOv5

为了提高工程车司机驾驶行为识别的精度和效率,本系统使用YOLOv5实现机车司机操作行为识别与视频监控,保证行车安全。本系统可实现疲劳驾驶监测、分心驾驶监测等多种不规范驾驶行为的检测,并且支持多路视频监测,提高监测效率。首先,利用车载摄像头采集司机驾驶状态视频流,传入检测模型;其次,通过网络模型进行实时监测,若出现不规范驾驶行为将发出警报并储存违规信息。为提高检测精度,本系统可通过人机交互界面自行绘制异形检测框,降低复杂背景对检测的不良影响。经过试验测试,该系统的损失在0.02左右,单类别检测准确率达到了92.5%,平均精度达到83.6%以上,能有效检测疲劳驾驶和驾驶时的不良行为,并及时发出语音警报提醒驾驶员。该系统将在减少工程事故、提高作业安全水平上发挥重要作用。

In order to improve the recognition accuracy and efficiency of driving behavior of engineering vehicle drivers,this system uses YOLOv5 to realize the recognition and video monitoring of locomotive driver operation behavior to ensure driving safety.This system can detect various irregular driving behaviors such as fatigue driving surveillance and distracted driving surveillance,and support multi-channel video surveillance to improve surveillance efficiency.Firstly,the video stream of driver’s driving state is collected by the vehicle camera and transmitted to the detection model;Secondly,real time surveillance is carried out through the network model.If irregular driving behavior occurs,an alarm will be issued and illegal information will be stored.In order to improve the detection accuracy,the system can draw a special-shaped detection frame by human-computer interaction interface,and reduce the adverse effects of complex background on detection.After experimental test,the loss of the system is about 0.02,the accuracy of single category detection reaches 92.5%,and the average accuracy reaches more than 83.6%.It can effectively detect the bad behavior of fatigue driving and driving,and send out voice alarm to remind drivers in time.This system will play an important role in reducing engineering accidents and improving operation safety.